RIS profile design method based on sparse arbitrary array and successive approximation technology

Abstract

The invention discloses an RIS profile design method based on a sparse arbitrary array and a successive approximation technology, and the method comprises the steps: deploying the sparse arbitrary array at an RIS sensing module, and reducing the number of radio frequency links; decoupling azimuth angle and pitch angle parameters by using a continuous approximation technology, and designing a low-complexity two-dimensional estimation algorithm; the RIS phase matrix is dynamically optimized based on the sensing result, and the communication and sensing performance is improved. Simulation shows that compared with a traditional 2D-MUSIC method, the scheme provided by the invention has the advantages that the RMSE (angle estimation error) and the spectral efficiency are obviously improved, and the calculation time consumption is reduced by more than 60%. A sparse arbitrary array architecture and a successive approximation technology are combined to solve the problems of high hardware cost and high calculation complexity in a semi-passive RIS-assisted ISAC system.

Description

Technical Field

[0001] The present invention relates to wireless communication technology methods, and in particular to a RIS profile design method based on sparse arbitrary arrays and continuous approximation technology. The method is applicable to a semi-passive RIS-assisted integrated communication and perception (ISAC) system, and is particularly targeted at low-complexity channel estimation and dynamic beamforming optimization in the millimeter wave frequency band. Background Art

[0002] As wireless communication technology evolves toward higher frequencies and higher integration, reconfigurable smart surfaces (RIS), a core technology for 6G networks, significantly improve the coverage and transmission efficiency of communication systems by dynamically manipulating the propagation characteristics of electromagnetic waves. However, existing semi-passive RIS-assisted ISAC systems face multiple technical challenges in practical deployment, severely limiting their potential for large-scale application.

[0003] Hardware cost and energy consumption are issues. Sensing modules in traditional semi-passive RIS architectures require densely packed sensors to achieve high-precision parameter estimation. For example, in a typical large-scale RIS system using a uniform array layout, the number of sensors increases quadratically with the array size, resulting in a large number of RF links. This not only significantly increases hardware manufacturing costs but also leads to high energy consumption.

[0004] The efficiency bottleneck of high-dimensional parameter estimation. ISAC systems require real-time acquisition of target azimuth and elevation information to optimize RIS phase configuration. Existing mainstream methods, such as the two-dimensional multiple signal classification (MUSIC) algorithm, require a global search in the two-dimensional angular space, resulting in extremely high computational complexity.

[0005] Performance degradation caused by grid mismatch. Existing compressed sensing-based solutions typically assume that target angles lie on a predefined, discrete grid. However, in real-world scenarios, the true target angle is often continuously distributed. When the angle deviates from the grid center, the reconstruction error accumulates linearly with the deviation distance. This grid mismatch not only reduces angle estimation accuracy but also causes RIS beamforming direction deviation, leading to degraded communication link quality and reduced perception sensitivity.

[0006] To address the above issues, some optimization strategies have been proposed, but they still have significant flaws. Sparse array design reduces the number of sensors and lowers costs, but random sparse layout introduces grating lobes in the directional pattern, reducing angular resolution and exacerbating multipath interference. Subspace decomposition methods reduce dimensionality and computational complexity, but they do not address the coupling problem between azimuth and elevation angles, requiring joint optimization, resulting in limited efficiency gains. Iterative optimization strategies, such as gradient descent, can adjust grid points to approximate the true angle, but require multiple iterations for convergence, making it difficult to meet low-latency requirements. Furthermore, the iteration process consumes significant computational resources, limiting its application in resource-constrained devices.

[0007] In summary, existing solutions often independently design perception and communication modules, lacking in-depth exploration of ISAC synergy gains. Existing semi-passive RIS-assisted ISAC systems face significant challenges in hardware architecture, computational efficiency, estimation accuracy, and functional synergy. An innovative solution is urgently needed that can deeply integrate hardware design and algorithm optimization to overcome the cost-performance trade-off and provide technical support for the efficient deployment of 6G networks. Summary of the Invention

[0008] Purpose of the invention: To address the problems of high hardware cost, complex two-dimensional parameter estimation calculation, and large grid mismatch error in semi-passive RIS-assisted ISAC systems, the present invention aims to propose an adaptive RIS profile design method based on sparse arbitrary arrays and continuous approximation technology:

[0009] Technical solution: The RIS profile design method based on sparse arbitrary array and continuous approximation technology includes the following steps:

[0010] Step (1) deploying a sparse arbitrary array architecture in the sensor module of the RIS, configuring a specified number of sensors in the horizontal direction and the vertical direction respectively, with the sensor spacing meeting the minimum half-wavelength constraint and the maximum spacing not exceeding a preset threshold, so as to reduce the number of RF links and reduce hardware costs;

[0011] Step (2) constructs a spatial angle parameter estimation problem based on the received signal model, maps the azimuth and elevation parameters into frequency-dependent horizontal and vertical angles, and divides the two-dimensional parameter space into discretized sub-regions through the continuous approximation technique;

[0012] Step (3) performs singular value decomposition and dimensionality reduction processing on the steering vector of each sub-region, retains the main eigenvector to construct a low-rank subspace, and realizes the decoupled estimation of azimuth and elevation angles;

[0013] Step (4): Use the improved MUSIC algorithm to search for the angle parameter corresponding to the minimum eigenvalue in the subspace after dimensionality reduction, and obtain the estimated values of the horizontal angle and the vertical angle respectively;

[0014] Step (5) dynamically updates the phase configuration of the RIS passive reflection module based on the angle estimation result, and optimizes the signal-to-noise ratio and spectrum efficiency of the communication link by adjusting the phase offset of each reflection unit.

[0015] Through the above process, a semi-passive RIS-assisted ISAC system model was constructed, combining sparse array deployment with successive approximation techniques to reduce hardware cost and computational complexity. The two-dimensional angle estimation problem was decomposed into independent subproblems for horizontal and vertical angles, achieving high-precision parameter estimation through singular value decomposition and an improved MUSIC algorithm. The phase configuration of the RIS passive reflector module was optimized based on dynamically updated angle parameters to generate directional beams and enhance communication link quality. Finally, through threshold pruning and group update strategies, a low-complexity algorithm was used to achieve channel reconstruction and improve system performance.

[0016] Furthermore, in step (1), the minimum distance between sensors is λ c / 2, the maximum spacing does not exceed (2min(N s，h , N s，v )-1)λ c / 2.

[0017] Furthermore, in step (1), the RIS sensing module receives the reflected signal from the UE to achieve target perception, and the UE's received signal model is:

[0018]

[0019] where h ru =α ru a(ψ ru ) is the channel vector from RIS to UE, is the path loss coefficient, d ru is the distance between RIS and UE, a(ψ ru ) is the steering vector of the RIS array, and its nth element is H br =α br a(ψ br，re )a T (ψ br,tr ) is the channel matrix from BS to RIS, a(ψ br，tr ) and a( ψbr,re ) are the steering vectors of the BS transmitter and the RIS receiver, Ω t =diag([ e jφt,1 ..., e jφt，N ] T ) is the phase configuration matrix of RIS, φ t，n represents the phase offset of the nth passive component at time slot t.

[0020] Furthermore, in step (1), the reflected signal received by the RIS sensor module is:

[0021] y r，t =αrus b(ψ ru )a T (ψ ru )Ω t H br fs t +n r，t

[0022] in, is the composite attenuation coefficient, β res is the target radar cross section, b(ψ ru ) is the steering vector of the sensor module, and its nth element is q rs，n is the sensor position vector.

[0023] Furthermore, in step (2), by introducing the normalized space angle parameter φ xo =sinψ ru，e cosψ ru，a and φ zo =cosψ ru，e , the steering vector of the sensor module is simplified to:

[0024]

[0025] Among them, ρ x，n =q x，n / λ c and ρ z，n =q z，n / λ c is the normalized position of the sensor, q x，n and q z，n is the XZ plane coordinate, and the continuous two-dimensional angle space A 2D ={(φ x ,φ z )|φ x ∈[-φ xb ,φ xb ],φ z ∈[-φ zb ,φ zb ]}divided into overlapping sub-regions Each sub-region corresponds to a discrete grid point By calculating the sub-region projection matrix:

[0026]

[0027] in, And extract the main eigenvectors through singular value decomposition (SVD) Construct low-dimensional subspace.

[0028] Furthermore, in step (4), an improved MUSIC algorithm is used to achieve angle estimation, and the specific steps are as follows:

[0029] First, calculate the sample covariance matrix, accumulate the received signals of J time slots, and calculate the sample covariance matrix:

[0030]

[0031] Among them, Y r，T，j =[y r，1，j， ..., y r，T，j ] is the received signal matrix of the jth time slot, and then the noise subspace is extracted. Perform eigendecomposition to obtain the signal subspace and noise subspace The noise subspace is composed of the eigenvector corresponding to the minimum eigenvalue. On this basis, a spectral function is constructed to perform angle estimation. The horizontal angle is estimated by minimizing the spectral function of the noise subspace projection:

[0032]

[0033] Similarly, estimate the pitch angle This step is achieved by reducing the dimensionality of the subspace V(φ x，s ) decouples the two-dimensional search into two one-dimensional searches, significantly reducing the computational complexity.

[0034] Furthermore, the step (5) is performed according to the estimated angle and Update the phase configuration of RIS passive components:

[0035]

[0036] in, and is the normalized position of the passive components.

[0037] In terms of hardware design, the RIS sensor modules are deployed using a sparse arbitrary array architecture. Sensors are arranged unevenly in the horizontal and vertical directions, with a minimum spacing of half a wavelength. This breaks the traditional periodic structure, suppresses grating lobes in the directional pattern, and increases spatial freedom. Furthermore, the maximum sensor spacing is limited to a preset threshold to avoid angular resolution degradation caused by excessive sparsity, ensuring accurate capture of target azimuth and elevation information even with limited hardware resources.

[0038] At the algorithmic level, a continuous approximation technique is proposed to address the efficiency bottleneck of traditional two-dimensional joint search. First, the continuous two-dimensional angular space is divided into multiple overlapping subregions, each associated with a discrete grid point and its neighborhood. Singular value decomposition is used to extract the principal eigenvectors of the signal covariance matrix within each subregion, constructing a low-dimensional signal subspace and decoupling the global two-dimensional search into a step-by-step one-dimensional search. A dynamic grid optimization mechanism is further introduced, using gradient descent to iteratively adjust the initial grid points, gradually approximating the true angle value and effectively eliminating grid mismatch errors.

[0039] Based on real-time angle estimation results, a dynamic RIS phase optimization strategy is designed. A phase mapping model is established between the reflector unit position and the target angle. The phase compensation for each unit is calculated based on the azimuth and elevation angles, generating a directional beam to enhance the signal strength in the target direction. A slot-level update mechanism, combined with a sliding window to fuse historical sensor data, suppresses transient noise interference and ensures the stability of the phase configuration. Furthermore, a threshold pruning strategy is introduced to evaluate the contribution of grid points to channel reconstruction in real time, dynamically removing redundant grid points to reduce computational complexity. Subcarriers are grouped and share a shared average wavelength dictionary to further reduce optimization complexity.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] Simulation results show that the proposed scheme significantly improves the RMSE and spectral efficiency compared with the traditional 2D-MUSIC method, and reduces the computational time by more than 60%. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the downlink transmission of the RIS-enabled ISAC system shows a schematic diagram of the semi-passive RIS deployment architecture. This figure illustrates the hardware layout of the proposed sparse arbitrary array RIS system. The central area consists of a 5×5 sparse sensor module. The sensors are distributed non-uniformly in the horizontal and vertical directions, with a minimum spacing of half a wavelength (λc / 2) and a maximum spacing not exceeding a preset threshold (e.g., 3λc). Surrounding the periphery are 64×64 passive reflective elements, forming a rectangular distribution excluding the central area. This design significantly reduces hardware cost by reducing the number of RF chains (from 256 in traditional solutions to 25). The sparse layout also suppresses grating lobes in the radiation pattern and improves angular resolution.

[0043] Figure 2 To support the adaptive RIS profile design and communication strategy of ISAC, the proposed dynamic optimization process is described. The left side shows the two-dimensional angle estimation process of the sparse sensing module, and the right side shows the real-time update mechanism of the RIS phase configuration. The arrows and block diagrams show the complete link from signal reception and parameter estimation to phase optimization, highlighting the core concept of ISAC functional collaboration.

[0044] Figure 3 The performance of ISAC under different RIS profile design strategies varies with the number of time slots. This figure shows the performance comparison of ISAC under different time slots. This figure compares the proposed method (CADRE) with the traditional 2D-MUSIC algorithm, SOMP algorithm, and random phase scheme in azimuth (ψ ru，a ) and pitch angle (ψ ru，e ) in the estimation of the root mean square error (RMSE) and the communication spectrum efficiency (β com ) changes with the time slot number, Figure a shows the ψ s，a Perceptual performance,The figure shows that as the number of time slots increases from 1 to 20, the RMSE of CADRE quickly converges to below 0.1° within 10 time slots, while 2D-MUSIC requires 20 time slots to achieve the same accuracy. In the initial stage (the number of time slots ≤ 5), the error of CADRE decreases 3 times faster than the traditional method. Figure b is ψ s,e The figure shows the perception performance of CADRE. The RMSE trend of the elevation angle is consistent with that of the azimuth angle, converging to 0.15° within 10 time slots. However, due to the grid mismatch problem, the RMSE of SOMP and random phase schemes are always higher than 1.5°. Figure c shows the communication performance. The figure shows that in terms of spectrum efficiency, CADRE reaches 12.5bps / Hz at 20 time slots, close to the theoretical upper limit, while the performance of 2D-MUSIC and SOMP drops to 5.2bps / Hz and 3.8bps / Hz respectively due to the beam tilt effect and grid mismatch problem. In addition, the spectrum efficiency of the random phase scheme is always lower than 4bps / Hz, verifying the necessity of dynamic phase optimization.

[0045] Figure 4 The figure shows the change of root mean square error and time cost with transmission power, showing the impact of transmission power on RMSE and computational time. This figure analyzes the performance difference between CADRE and the comparison algorithm when the transmission power (Pt) increases from 5dBm to 50dBm. Figure a is the perceptual performance. It shows that when Pt≥≥20dBm, the RMSE of CADRE is stable within 0.3°, which is comparable to the accuracy of 2D-MUSIC. However, due to the grid mismatch error, the RMSE of SSD and SOMP only decreases slightly with increasing power (from 2.1° to 1.5°); Figure b is the computational complexity. It shows that when Pt=30dBm, CADRE takes only 40% of 2D-MUSIC, and the time consumption increases slowly with increasing power, while the time consumption of 2D-MUSIC increases exponentially (from 15ms to 85ms), verifying the effectiveness of the threshold pruning and group update strategy. DETAILED DESCRIPTION

[0046] This paper proposes a hardware-efficient and low-complexity adaptive RIS profile design strategy for a typical semi-passive RIS-assisted ISAC system. Specifically, to reduce hardware cost while maintaining high estimation accuracy, the RIS sensor modules are deployed in a sparse arbitrary array. Based on this array configuration, an efficient parameter estimation algorithm based on successive approximation is designed. In combination with the sensing results, a dynamic profile design method supporting ISAC is proposed to optimize the phase configuration of the passive RIS modules.

[0047] The RIS profile design method based on sparse arbitrary array and continuous approximation technology of the present invention comprises the following steps:

[0048] Step (1) deploying a sparse arbitrary array architecture in the sensor module of the RIS, configuring a specified number of sensors in the horizontal direction and the vertical direction respectively, with the sensor spacing meeting the minimum half-wavelength constraint and the maximum spacing not exceeding a preset threshold, so as to reduce the number of RF links and reduce hardware costs;

[0049] Step (2) constructs a spatial angle parameter estimation problem based on the received signal model, maps the azimuth and elevation parameters into frequency-dependent horizontal and vertical angles, and divides the two-dimensional parameter space into discretized sub-regions through the continuous approximation technique;

[0050] Step (3) performs singular value decomposition and dimensionality reduction processing on the steering vector of each sub-region, retains the main eigenvector to construct a low-rank subspace, and realizes the decoupled estimation of azimuth and elevation angles;

[0051] Step (4): Use the improved MUSIC algorithm to search for the angle parameter corresponding to the minimum eigenvalue in the subspace after dimensionality reduction, and obtain the estimated values of the horizontal angle and the vertical angle respectively;

[0052] Step (5) dynamically updates the phase configuration of the RIS passive reflection module based on the angle estimation result, and optimizes the signal-to-noise ratio and spectrum efficiency of the communication link by adjusting the phase offset of each reflection unit.

[0053] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] First, build a system model:

[0055] Step 1. Establishment of system model

[0056] In step (1.1), we consider a semi-passive RIS-assisted ISAC downlink transmission system, where the base station (BS) is equipped with a uniform linear array (ULA) with Mt antennas. The RIS consists of an active sensing module and a passive reflection module. The BS transmits communication signals to a single-antenna user equipment (UE) via the RIS reflection path. At the same time, the RIS sensing module receives the reflected signal from the UE to achieve target perception. The UE's received signal model is:

[0057]

[0058] where h ru =α ru a(ψ ru ) is the channel vector from RIS to UE, is the path loss coefficient, d ru is the distance between RIS and UE, a(ψ ru ) is the steering vector of the RIS array, and its nth element is H br =α br a(ψ br，re )a T (ψ br,tr ) is the channel matrix from BS to RIS, a(ψ br，tr ) and a(ψ br,re ) are the steering vectors of the BS transmitter and the RIS receiver respectively. t =diag([e jφt，1 ..., e jφt，N ] T ) is the phase configuration matrix of RIS, φ t，n represents the phase offset of the nth passive component at time slot t.

[0059] In step (1.2), the reflected signal received by the RIS sensor module is:

[0060] y r，t =α rus b(ψ ru )a T (ψ ru )Ω t H br fs t +n r，t

[0061] in, is the composite attenuation coefficient, β res is the target radar cross section. b(ψ ru ) is the steering vector of the sensor module, and its nth element is q rs，nis the sensor position vector. It is worth noting that here we use a sparse array design, and the sensor module uses N s =N s，h ×N s，v Sparse arbitrary array, the minimum sensor spacing is λ c / 2, the maximum spacing does not exceed (2min(N s，h , N s，v )-1)λ c / 2 to reduce the number of RF chains (for example, only 5×5 sensors are needed in a 64×64 RIS).

[0062] The following is the RIS profile design method proposed by the present invention:

[0063] Step 2. RIS profile design based on successive approximations

[0064] Step (2.1), by introducing the normalized space angle parameter φ xo =sinψ ru，e cosψ ru，a and φ zo =cpsψ ru,e , the steering vector of the sensor module is simplified to:

[0065]

[0066] Among them, ρ x，n =q x，n / λ c and ρ z，n =q z，n / λ c is the normalized position of the sensor, q x，n and q z，n is the XZ plane coordinate. Step (2.2), the continuous two-dimensional angle space A 2D ={(φ x ,φ z )|φ x ∈[-φ xb ,φ xb ],φ z ∈[-φ zb ,φ zb ]}divided into overlapping sub-regions Each sub-region corresponds to a discrete grid point By calculating the sub-region projection matrix:

[0067]

[0068] in, And extract the main eigenvectors through singular value decomposition (SVD) Construct low-dimensional subspace.

[0069] In step (2.3), we use the improved MUSIC algorithm to achieve angle estimation. The specific steps are as follows.

[0070] First, calculate the sample covariance matrix. Accumulate the received signal of J time slots and calculate the sample covariance matrix:

[0071]

[0072] Among them, Y r，T，j =[y r，1， j, ..., y r,T，j ] is the received signal matrix of the jth time slot.

[0073] Then the noise subspace is extracted. Perform eigendecomposition to obtain the signal subspace and noise subspace The noise subspace consists of the eigenvectors corresponding to the minimum eigenvalues.

[0074] On this basis, the spectrum function is constructed to estimate the angle. The horizontal angle is estimated by minimizing the spectrum function of the noise subspace projection:

[0075]

[0076] Similarly, estimate the pitch angle This step is achieved by reducing the dimensionality of the subspace V(φ x，s ) decouples the two-dimensional search into two one-dimensional searches, significantly reducing the computational complexity.

[0077] Step (2.4), dynamic phase optimization. According to the estimated angle and Update the phase configuration of RIS passive components:

[0078]

[0079] in, and is the normalized position of the passive components.

[0080] Step 3. Computational efficiency optimization and performance verification

[0081] Step (3.1), threshold pruning and subcarrier grouping. We adopt the threshold pruning strategy, that is, in each iteration, if the covariance contribution of a grid point is Below the threshold α th =0.1max(λ), then remove the grid point to reduce the amount of calculation. In addition, we divide the M subcarriers into P groups (e.g., P=8), and each group shares the average wavelength dictionary The computational complexity is reduced from O(M) to O(P).

[0082] In step (3.2), in the simulation verification, we conduct performance evaluation based on the following parameter configuration: the RIS array size is 64×64, the sensor module adopts a 5×5 sparse arbitrary array, the base station (BS) is configured with 64 antennas, and the carrier frequency is 30GHz (corresponding to the wavelength λ c =1cm), and the noise power spectral density is -174dBm / Hz. Through experimental analysis, the proposed method performs well in multiple key indicators. First, when the signal-to-noise ratio (SNR) is 20dB, the root mean square error (RMSE) of the angle estimation is less than 0.5°, which is 3 times more accurate than the traditional 2D-MUSIC algorithm; secondly, under the same SNR, the spectral efficiency of the communication link reaches 12.5bps / Hz, which is 2.3 times higher than the existing method; in addition, the computational time is significantly reduced, and the phase update can be completed in only 30% of the time of the traditional algorithm, meeting the stringent real-time requirements of the millimeter wave channel. These results verify the comprehensive advantages of sparse array design and continuous approximation technology in hardware cost, computational efficiency and collaborative optimization of perception and communication.

Claims

1. A RIS profile design method based on sparse arbitrary array and continuous approximation technology, characterized in that: The steps include: Step (1) deploying a sparse arbitrary array architecture in the sensor module of the RIS, configuring a specified number of sensors in the horizontal direction and the vertical direction respectively, with the sensor spacing meeting the minimum half-wavelength constraint and the maximum spacing not exceeding a preset threshold, so as to reduce the number of RF links and reduce hardware costs; Step (2) constructs a spatial angle parameter estimation problem based on the received signal model, maps the azimuth and elevation parameters into frequency-dependent horizontal and vertical angles, and divides the two-dimensional parameter space into discretized sub-regions through the continuous approximation technique; Step (3) performs singular value decomposition and dimensionality reduction processing on the steering vector of each sub-region, retains the main eigenvector to construct a low-rank subspace, and realizes the decoupled estimation of azimuth and elevation angles; Step (4): Use the improved MUSIC algorithm to search for the angle parameter corresponding to the minimum eigenvalue in the subspace after dimensionality reduction, and obtain the estimated values of the horizontal angle and the vertical angle respectively; Step (5) dynamically updates the phase configuration of the RIS passive reflection module based on the angle estimation result, and optimizes the signal-to-noise ratio and spectrum efficiency of the communication link by adjusting the phase offset of each reflection unit.

2. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: In step (1), the minimum distance between sensors is λ c / 2, the maximum spacing does not exceed (2min(N s，h , N s，v )-1)λ c / 2.

3. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: In step (1), the RIS sensing module receives the reflected signal from the UE to achieve target perception. The UE's received signal model is: where h ru =α ru a(ψ ru ) is the channel vector from RIS to UE, is the path loss coefficient, d ru is the distance between RIS and UE, a(ψ ru ) is the steering vector of the RIS array, and its nth element is H br =α br a(ψ br，re )a T (ψ br，tr ) is the channel matrix from BS to RIS, a(ψ br，tr ) and a(ψ br，re ) are the steering vectors of the BS transmitter and the RIS receiver, Ω t =diag([e jφt，1 ,...,e jφt，N ] T ) is the phase configuration matrix of RIS, φ t，n represents the phase offset of the nth passive component at time slot t.

4. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: In step (1), the reflected signal received by the RIS sensor module is: y r，t =a rus b(ψ ru )a T (ψ ru )Oh t H br fs t +n r，t in, is the composite attenuation coefficient, β res is the target radar cross section, b(ψ ru ) is the steering vector of the sensor module, and its nth element is q rs，n is the sensor position vector.

5. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: In step (2), by introducing the normalized space angle parameter φ xo =sinψ ru，e cosψ ru，a and φ zo =cosψ ru，e , the steering vector of the sensor module is simplified to: Among them, ρ x，n =q x，n / λ c and ρ z，n =q z，n / λ c is the normalized position of the sensor, q x，n and q z，n is the XZ plane coordinate, The continuous two-dimensional angle space Divide into overlapping sub-regions Each sub-region corresponds to a discrete grid point By calculating the sub-region projection matrix: in, And extract the main eigenvectors through singular value decomposition (SVD) Construct low-dimensional subspace.

6. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: In step (4), the improved MUSIC algorithm is used to achieve angle estimation, and the specific steps are as follows: First, calculate the sample covariance matrix, accumulate the received signals of J time slots, and calculate the sample covariance matrix: Among them, Y r，T，j =[y r，1，j ,...,y r，T，j ] is the received signal matrix of the jth time slot, and then the noise subspace is extracted. Perform eigendecomposition to obtain the signal subspace and noise subspace The noise subspace is composed of the eigenvector corresponding to the minimum eigenvalue. On this basis, a spectral function is constructed to perform angle estimation. The horizontal angle is estimated by minimizing the spectral function of the noise subspace projection: Similarly, estimate the pitch angle This step is achieved by reducing the dimensionality of the subspace V(φ x，s ) decouples the two-dimensional search into two one-dimensional searches, significantly reducing the computational complexity.

7. The RIS profile design method based on sparse arbitrary array and continuous approximation technology according to claim 1, characterized in that: The step (5) is based on the estimated angle and Update the phase configuration of RIS passive components: in, and is the normalized position of the passive components.

Images